Speech recognition
Summary Description Newswire's 5 Minute Guides looks at controlling a computer with your vocal cords.
Author

Publication

Roullas Top10 Simon Vandore

Newswire
No


Editorial InformationArticle Location
Article Topic 5 Minute Guides
Story Order
Story Group 001119
Post Date 13/11/2000 08:02 AM Status Posted Entered by Simon Vandore on 10/11/2000 03:51 PM


ImagesLead Picture


Heading Image




Content
Introduction


Body
How can a computer understand your voice?
When you use a speech recognition program, the sounds of your voice are picked up by a microphone attached to your computer and turned into digital information. The software contains a database of sounds and whole words, along with typical patterns of speech, which it matches with your digitised voice and displays as text on the screen.
What is speech recognition technology used for?
People do most of their communicating with speech, but are confined to hitting little plastic keys whenever they want to interact with machines. Speech recognition aims to overcome this problem.
The most obvious use is to dictate instead of type into a word processor. Most of the commercial packages are aimed primarily at this activity, but many support voice control of your computer's other functions. There are also many potential industrial applications. The biggest consumers of speech recognition software are those with a reputation for hard-to-read handwriting -- doctors, lawyers and business people who would otherwise rely on dictation with human transcribers. Users with physical disabilities are another obvious market. There are also users who have suffered from repetitive strain injury (RSI) from typing, and still others who simply want to avoid that risk.
What are the limitations and how can they be overcome?
A few years ago, users of speech recognition programs had to speak ... very ... slowly ... into their microphones. Today, several packages are available that handle a normal conversational pace. When you first install a speech recognition program, you are asked to read it some text so that the program can get used to your voice and pronunciation. This improves accuracy (often to over 95%), while using a version of the software built for your national accent can help a great deal.
Efficient speech recognition demands a great deal of processing power and RAM. In an ideal situation, the sound database is loaded entirely into RAM so that the program doesn't need to search for speech patterns on a slower hard drive. In recent years, a new generation of digital 'array' microphones has become available to improve sound capture, which can be distorted by background noise such as a busy office or family home.
Why do people still use a mouse and a keyboard?
The trouble is that computers are still designed for push-button control. Some things are much easier to achieve with a keypress than a series of spoken words, especially when many words sound the same, such as "pear" and "pair". The solution is artificial intelligence which can recognise contextual use of words and guess when you really mean "pear", but this requires even more number-crunching power. The average office worker's typing speed is well under 50 words per minute, while people talk at 80 to 150 words per minute, so there are plenty of opportunities for development.
What's the state of play in the speech recognition industry?
The three biggest providers of speech recognition software for PCs are Dragon Systems, IBM and Lernout & Hauspie. Dragon products hold over 60% market share, but the company was purchased from Drs Jim and Janet Baker by Lernout & Hauspie in March 2000 to create a dominant player. However, this week Lernout & Hauspie reported financial irregularities and its Belgian founders were forced to resign and Nasdaq trading in the company's shares was suspended.
There are also many companies developing industry-specific speech recognition solutions, especially those used by call centres. For example, Sydney's Combined Taxis has ordered a telephone reservation system which will recognise specifically Australian speech patterns. Even Microsoft is planning to incorporate speech recognition into upcoming versions of Office.


Related MaterialsRelated Articles


Related Links




Bulletin Summary5 Minute Guides: Speech recognition
Newswire's 5 Minute Guides looks at controlling a computer with your vocal cords.

WAP Summary

Cross-Publishing InformationShort Headline
5 Minute Guides: Speech recognition

Clipping Information

Corporate IT No This field should be marked 'Yes' for any story of interest to corporate readers
CIT Lead No Newswire Lead No Section Lead No (These fields are controlled by all those handy buttons and agents)